A near-surface high-density sensor network with 146 sites in the central urban area of Xiangyang City，Hubei Province，China，was established to collect high-resolution data on PM_2.5，PM₁₀，wind speed，and wind direction to solve the problem of difficulty in identifying the dispersion patterns of urban-scale particulate-matter （PM） and the lack of planning feasibility. On this basis，a method of identifying pollution that balances the interpretability of mechanism and the efficiency of computation was proposed.Multi-scale continuous characterization of affecting factors including industrial emissions，fugitive dust，and blue-green spaces was conducted under the constraints of Gaussian diffusion mechanism.A dual-precision nested strategy was proposed to screen key factors.The geographically weighted regression （GWR） was used to analyze the spatial heterogeneity effects of various factors on particulate matter.The results showed that industrial sources clustered in a 1.2-4 km area around thermal power plants and industrial parks，and intensified along the “park-logistics corridor” pathways.The construction sites and freight logistics formed banded high-impact zones along the major transportation axes，and the effect of logistics activities on PM₁₀ was significantly higher than that on PM_2.5.The blue spaces formed about 2-kilometer-wide mitigation belts at the coupling point between the area with low-intensity of development and the main ventilation corridors，while the high-intensity of development weakened its effect.The green spaces generally reduced PM，especially PM₁₀，but short-term benefits was offset by construction disturbances.Based on this，a translation path with a “zone-corridor-node” priority scheme，targeted greening configurations along logistics corridors and park edges，pocket-scale multilayer planting at key nodes，and continuity of low-density layouts and ventilation corridors for planning practice was proposed.It is indicated that the established method achieves a better balance between the interpretability，accuracy，and the computational power compared to the traditional methods of spatial statistics，and can identify key segments of homologous heterogeneity and exogenous overlap.